Interference targets opinion

For years, election security meant locked server rooms, paper backups, and making sure nobody could change a vote after it was cast. The newer problem is cheaper and messier: flood voters with false stories before they ever reach the ballot box. (governing.com) Federal agencies said this shift out loud during the 2024 election cycle. In an October 18, 2024 warning, the Federal Bureau of Investigation and the Cybersecurity and Infrastructure Security Agency said they had no information that malicious cyber activity had changed ballots or disrupted counting, but they did see foreign actors pushing false claims to undermine confidence in the process. (ic3.gov) That is a different kind of attack. Instead of breaking into the vote-counting machine, the attacker tries to break the voter’s picture of reality with fake videos, fake audio, spoofed websites, text messages, and fake local-news style posts. (ic3.gov) Generative artificial intelligence made that job faster in 2024. The same Federal Bureau of Investigation and Cybersecurity and Infrastructure Security Agency alert said these tools lowered the barrier for foreign actors to produce synthetic pictures, deepfake video, deepfake audio, and inauthentic articles at greater speed and scale. (ic3.gov) The clearest American example was New Hampshire’s presidential primary on January 23, 2024. Voters got robocalls two days earlier that used an artificial-intelligence clone of President Joe Biden’s voice to tell Democrats to stay home and “save” their vote for November. (doj.nh.gov) Washington reacted fast after that call. On February 8, 2024, the Federal Communications Commission ruled that artificial-intelligence-generated voices in robocalls are “artificial” under the Telephone Consumer Protection Act, giving state attorneys general and federal regulators a clearer path to go after voice-cloning scams. (fcc.gov) Even when a fake does not stop a single ballot from being counted, it can still do damage. The Brennan Center wrote in March 2025 that the bigger effect of deepfakes is erosion of trust: voters become less sure what is real, more polarized, and more willing to dismiss genuine evidence as fake. (brennancenter.org) That is why the next layer of defense is not only cybersecurity but provenance. Provenance is the record of where a photo, video, or audio clip came from and what was changed, like a shipping label plus a repair log for a package moving across the internet. (openai.com) One of the main standards here is from the Coalition for Content Provenance and Authenticity. Its system adds “Content Credentials,” which the group describes as a kind of nutrition label for digital media that shows origin and edits. (c2pa.org) OpenAI said in May 2024 that it was adding Coalition for Content Provenance and Authenticity metadata to image tools and that cryptographically signed metadata avoids the false-positive problem that comes with trying to guess whether content is machine-made after the fact. In plain terms, a signed label attached at creation is more useful than a detector making a probabilistic guess after a clip has already gone viral. (openai.com) Public agencies started building around that reality during the last national election. On October 28, 2024, the Cybersecurity and Infrastructure Security Agency launched a central election-threat page inside its Protect 2024 program to collect joint statements, threat updates, and public warnings for more than 8,000 election jurisdictions. (cisa.gov) So the practical change for 2026 is simple. Campaigns, election offices, and newsrooms still need hardened systems and paper trails, but they also need workflows for verifying media, checking source records, flagging coordinated influence campaigns, and answering false claims before those claims harden into voter memory. (governing.com)